The Cancer Genome Atlas (TCGA)

Assay and analyze cancer samples

TCGA Data Coordination

The TCGA Pan-Cancer Initiative

TCGA Pan-Cancer Initiative

Focus on TCGA Pan-Cancer Initiative

Nature Genetics October 2013

TCGA analysis

Two fundamental observations

  • Intra-cancer heterogeneity

    Tumors originating in the same tissue vary substantially in genomic alterations.



  • Cross-cancer similarity

    Similar patterns of genomic alteration are observed in tumors from different tissues of origin.

Title of this paper

TL;DR

  • \(3299\) tumors across \(12\) different cancer types
  • Integrated multiple types of alterations (genomic & epigenomic)
  • Hierarchical stratification approach to obtain clusters of tumors
  • Observed two major clusters
    • M class: primarily somatic mutation
    • C class: primarily copy number alterations
  • Inverse relationship between # of copy number alterations & # of somatic mutations (when averaged over 12 cancer types)
  • Oncogenic signatures were used to derive the oncogenic pathways
  • Nominated therapeutically actionable targets across tumor types

\(3299\) tumors, \(12\) different cancer types

PANCAN12 dataset

Samples

Data reduction

Data processing flowchart


Integrated multiple types of alterations

Copy number aberrations (CNAs)

CNAs 1

Integrated multiple types of alterations

Copy number aberrations (CNAs)

CNAs 2

Integrated multiple types of alterations

Somatic mutations (SNVs)

SNVs 1

Integrated multiple types of alterations

Somatic mutations (SNVs)

SNVs 2

Integrated multiple types of alterations

DNA methylation

Methyl 1

Integrated multiple types of alterations

DNA methylation

Methyl 2

Data for further analysis

Selected functional events (SFEs)

SFEs


  • treated as binary variables (occured or not occured)
  • network that connects samples to alterations

Bipartite network analysis

samples connected to alterations

Bipartite graph

Bipartite network analysis

Find optimal modules

Bipartite graph

Data for further analysis

Find optimal modules recursively

Bipartite graph

Hierarchical stratification to obtain clusters

Network modularity

Cluster example 2

Hierarchical stratification to obtain clusters of tumors

Equations

Observed two major clusters

M(utations) vs C(opy) number alterations

Fig 2a

Features correspond to identified classes

Fig 2b

Inverse relationship b/t CNA and SNV

Cancer hyperbol[a|e]

Fig 2b

Inverse relationship b/t CNA and SNV

Cancer hyperbol[a|e]

Fig S3

Validation PANCAN18 dataset

Fig S4

Signatures in the M class

Figure 3M

Signatures in the C class

Figure 3C

Therapeutically actionable targets

Four well studied pathways across tumor types

Figure 4

Therapeutically actionable targets

across tumor types

Figure 5

Conclusions

  • Two major classes of tumors with different oncogenic drivers (mutation and copy number change)
  • Argues for tissue-independent classification and treatment of tumors
  • Almost all solid tumors, many other tumor types not covered
  • Though the methods should get better with more data

Title of this paper



An Analysis is a new analysis of existing data (typically large genomic, transcriptomic or proteomic data sets from arrays or other high-throughput platforms) or new data obtained in a comparative analysis of technologies that lead to novel and arresting conclusions of importance to a broad audience.

Sage Bionetworks

Synapse



Thanks!